Semantic Relations

A semantic relationship of a concept to another concept is represented by an edge between the concept nodes in question. The edge label indicates the type of edge and thus the type of relationship, as illustrated in Figure 24.

Edges are directed whenever the semantic relations represented by them are not symmetrical, i.e., the nodes at the different ends of an edge are to be interpreted differently with respect to their place in the relation, e.g., as upper or lower term (is-a), as part or whole (has-part). The direction can be defined arbitrarily for an edge type (e.g. for is-a from the subterm to the superterm), but is binding for all expressions of the edge type.

We have so far distinguished two types of nodes: concept nodes and property nodes, and we have distinguished two types of edges: edges for semantic relations as well as for the assignment of properties to concepts.

In the literature, this distinction is not always made (e.g., (Rich 83) and sometimes all nodes in a net are considered as concept or object nodes. However, since a clear and meaningful distinction can be made between concepts and properties (cf. chap.1.4), we want to be able to reflect this distinction in a semantic net as well.

In order for a semantic network to be correctly interpreted (both by a human observer and by a program), the type of each node and edge must be recognizable from a representation, because different meanings are associated with different types. Thus, based on the semantics of property nodes, property edges that start from (rather than lead to) such a node are nonsensical and without interpretation. The same is true for relation edges leading to or from a property node (cf. with the notion of a property as introduced in chap.1.4). Such inadmissible constellations can be avoided only if one can always read from a representation whether it concerns in each case a property node or a concept node. Ideally, the representation system that implements a net-like representation model and is responsible for managing individual net representations performs corresponding permissibility checks (also called integrity monitoring). Property and concept nodes are thus to be introduced as representation constructs. In the graphical illustrations of semantic networks, we distinguish heath node types by different font types (cf. Fig.25). Edges leading to property nodes are always property edges and can be distinguished from relation edges, which always connect two concept nodes, on the basis of this criterion.